TreeSatAI-Time-Series Dataset Overview

Dataset Introduction

TreeSatAI-Time-Series is an extension of the TreeSatAI benchmark introduced by Ahlswede et al., focusing on Central European tree species classification. The dataset combines aerial, Sentinel‑1, and Sentinel‑2 multi‑sensor data and contains labels for 20 European tree species (15 genera) extracted from forest management records in Lower Saxony, Germany.

Dataset Features

• Time‑Series Extension: Unlike the original dataset, which provided only a single Sentinel‑1 and Sentinel‑2 image per plot, this version integrates all available Sentinel‑1 and Sentinel‑2 observations over a year, facilitating species discrimination.
• Temporal Alignment: For plots captured after 2017, the Sentinel time series are aligned with the aerial acquisition year. For earlier plots, 2017 is used to provide sufficient temporal context while assuming minimal forest change.

Dataset Composition

The collection comprises 50,381 plots (60 m × 60 m) across Germany and includes the following compressed folders:

• aerial: Original aerial data (0.2 m resolution) with RGB and NIR bands.
• sentinel: Single‑date Sentinel‑1 and Sentinel‑2 imagery covering either the plot (60 m) or a larger area (200 m).
• sentinel‑ts: Annual Sentinel‑1 and Sentinel‑2 time‑series.
• labels: Species and proportion labels per plot.
• geojson: Vector files with plot geometries.
• split: Train/validation/test plot splits.

Sentinel Time‑Series Data Format

Sentinel series are provided in HDF5 (.h5) containing:

• sen‑1‑asc‑data: Ascending‑orbit SAR backscatter (Tx2x6x6), channels VV, VH.
• sen‑1‑asc‑products: Ascending‑orbit product identifiers.
• sen‑1‑des‑data: Descending‑orbit SAR backscatter (Tx2x6x6), channels VV, VH.
• sen‑1‑des‑products: Descending‑orbit product identifiers.
• sen‑2‑data: Level‑2 BOA reflectance (Tx10x6x6), bands B02‑B12.
• sen‑2‑masks: Cloud masks (Tx2x6x6), snow and cloud probability.
• sen‑2‑products: Sentinel‑2 product identifiers.

Product names follow the official ESA naming conventions.

License

The dataset is released under the Creative Commons Attribution‑ShareAlike 4.0 International License.